From: Andries.Brouwer@cwi.nl

jpcartal@free.fr writes:

	I noticed that contrary to what was happening with 2.4.x kernel, suid
	root files don't loose their suid bit when they get overwritten by a
	normal user (see example below)

	Is this the intended behaviour or a bug ?

	Example :

	[root@localhost test]# chown root ~cartaljp/test/suid_test
	[root@localhost test]# chmod 4775 ~cartaljp/test/suid_test
	[root@localhost test]# exit
	[cartaljp@localhost test]$ cp /bin/ls suid_test
	[cartaljp@localhost test]$ ls -l
	total 72
	-rwsrwxr-x    1 root     cartaljp    67668 Sep 19 07:56 suid_test <-
	Suid bit is still set whereas with 2.4.x kernel it was reset.

Yes. Here 2.4 had the terrible code

     mode = (inode->i_mode & S_IXGRP)*(S_ISGID/S_IXGRP) | S_ISUID;

while 2.6 does things via notify_change().  However, in 2.6 notify_change()
does not allow removal of the SUID bit because you are not owner of the
file :-).  So, we have to convince inode_change_ok() to do it anyway.

At present you can set CONFIG_HUGETLB_PAGE and not CONFIG_HUGETLBFS, which
give the kernel low-level hugepage support, but no userspace API with which
to access it.

And with the recent hugetlbfs-accounting fix, the low-level code needs
functions which are in hugetlbfs, so CONFIG_HUGETLBFS=n,
CONFIG_HUGETLB_PAGE=y doesn't even link.

So we flip things around: CONFIG_HUGETLB_PAGE is set if CONFIG_HUGETLBFS is
set and CONFIG_HUGETLB_PAGE is unset if CONFIG_HUGETLBFS is unset.  The
CONFIG_HUGETLB_PAGE option hence disappears from the per-arch configuration
menus.

From: "Chen, Kenneth W" <kenneth.w.chen@intel.com>

We found a problem in hugetlbfs file system quota when using huge page
via mmap.  The mmap method in hugetlbfs_file_operation always takes
quota for every mmap even for pages that are already allocated on that
inode.  This results in taxing the same hugepage multiple times and
causing mmap to fail on existing file when quota mistakenly runs out.

- modules need blk_rq_prep_restart()  (Florian Schanda <ma1flfs@bath.ac.uk>)

- Remove unneeded inclusions from configs.c (Randy Dunlap)

- The agp_setup() ifdef was backwards (Stephen Hemminger <shemminger@osdl.org>)

- makefiles.txt typo fix (Paolo Ornati <ornati@despammed.com>)

From: Christoph Hellwig <hch@lst.de>

All drivers that compile on ppc with CONFIG_ISA set (= all but
some m68-only drivers), I looked at the compile warnings very
closely and there are no new warnings or even errors this time :)

drivers/char/Makefile needed to be reordered big time to keep
the intialization order the same.

From: Ethan Benson <erbenson@alaska.net>

Currently Linux allows the use of the utime() and utimes() syscalls on
immutable or append-only files, this is incorrect.

utime{s}() is not supposed to work if you lack write access to a file,
in the case of an immutable file NOBODY has write access.

for an append-only file it only makes sense to be able to update its
time to present, not the past.

I have checked BSD, and they implement the behavior I propose, for
immutable files utime() and utimes() fail.  for append-only files they
fail if the time argument is not NULL.

From: John Levon <levon@movementarian.org>

Not that many people are going to be using GCC 3.4 currently, but it
might help a bit to prevent compilation bugs like that just witnessed
in procfs.

(And it consolidates the check_gcc implementation nicely)

From: Christoph Hellwig <hch@lst.de>

Some junk which was needed before we did the big cleansweep for the sysrq
syncing stuff.

From: Christoph Hellwig <hch@lst.de>

Richard hasn't touched it for about a year and since then only bugfixes and
my changes to the kernel interface went in.  No one has stepped up to
maintain it and with udev we have a proper replacement now.

I will do some more work before 2.6 (mainly trying to implement Al's
refcounting scheme to avoid stale entries), but that's it.

From: "David S. Miller" <davem@redhat.com>

It's from Ingo Molnar.  ksoftirqd kicks in way too early, so do more work in
interrupt context before falling back.

We can probably sysctl this thing, that way everyone gets what they want
probably...

(has been in -mm since 2.5.71 and I haven't heard a peep).

From: Manfred Spraul <manfred@colorfullife.com>

- Remove cache_alloc_one_tail and cache_alloc_listfixup - there is no
  reason for their existance.

- Print a bit more debugging info when slab corruption is detected.

	Argh.  A couple of places where we needed ..._encode_dev() had
been lost in reordering the patchset - the most notable being ctty number in
/proc/<pid>/stat.  Fix follows:

This fixes several things including fixing the character ioctl polling
to use wait_event_interruptible_timeout() instead of
interruptible_sleep_on_timeout() and several other fixes.

These changes were already applied to 2.4.22.

into home.osdl.org:/home/torvalds/v2.5/linux

into home.osdl.org:/home/torvalds/v2.5/linux

From Szombathelyi György

This fixes a memory leak in the uniform cdrom layer.

From Szombathelyi György

This fixes a problem where the residual data count isn't being set
correctly.

Real conversion to 32bit dev_t.  Expansion to:
	* mknod() - 32
	* newstat() - 32 on 64bit platforms
	* stat64() - 32 on mips, 64 on everything else (mips has weird struct
stat64 and can't get more than 32 bits).  Note that right now the difference
is purely theoretical - we don't have internal values above 32 bits, so
huge_... vs. new_... only marks the places where 64bit conversion will need
extra work.
	* arch-dependent stat variants - depending on width available.
	* ustat et.al. - 32
	* filesystems that can handle 32 bits right now - 32
	* ext2 and ext3 - 32, with large dev_t inodes having 0 in the first
element of i_data[] (where we store dev_t value for small device numbers) and
keeping the value in the second element.
	* nfsd - 32; it can be driven to 64, but we'll get several issues with
NFSv2 support.
	* RAID - 32
	* devmapper - with v1 it's still 16 (nothing to do here), with v4 it's
64.
	* loop - 64
	* initramfs - 32
	* do_mounts code - 32.  Parts that scan devfs tree are using newstat()
on 64bit platforms and stat64() on the rest (IOW, the latest stat variant on
given platform).
	* old_valid_dev()/new_valid_dev() added where needed (stat variants,
mostly - we fail with -EOVERFLOW if values do not fit).

Starting the conversion:
	* internal dev_t made 32bit.
	* new helpers - new_encode_dev(), new_decode_dev(), huge_encode_dev(),
huge_decode_dev(), new_valid_dev().  They do encoding/decoding of 32bit and
64bit values; for now huge_... are aliases for new_... and new_valid_dev()
is always true.  We do 12:20 for 32bit; representation is compatible with
16bit one - we have major in bits 19--8 and minor in 31--20,7--0.  That's
what the userland sees; internally we have (major << 20)|minor, of course.
	* MKDEV(), MAJOR() and MINOR() updated.
	* several places used to handle Missed'em'V dev_t (14:18 split)
manually; that stuff had been taken into common helpers.
	Now we can start replacing old_... with new_... and huge_..., depending
on the width available.  MKDEV() callers should (for now) make sure that major
and minor are within 12:20.  That's what the next chunk will do.

	Added old_encode_dev() to assorted stat() variants.
	Fixed bug in s390 emulation on s390x: stat64() should never truncate
UID and GID.
	Ditto for i386 emulation on x86_64.
	Replaced dev_t in various struct stat with explicit integer type.
	Replaced __kernel_dev_t with __old_kernel_dev_t in dm-ioctl-v1.h
	Now we are free to change dev_t in any way we want - on all boundaries
we have explicit conversions.
	Took __kernel_dev_t definition to linux/types.h and changed it with
__u16.  We are ready to proceed to 32bit now.

	Added old_encode_dev()/old_decode_dev() in CODA code - on assignments
to/from ->va_rdev.

	Added old_encode_dev() in loop.c

	Changed sys_mknod() prototype to have unsigned int passed to it
instead of current dev_t.  Added old_decode_dev() in sys_mknod() and
made sure that its callers are passing it old_encode_dev(<value>)
	Switched sys_ustat() and its variants from dev_t to unsigned (and
added old_decode_dev()).
	Took care of assignments to ROOT_DEV - again, old_decode_dev().
	Late-boot search in devfs (call sys_newstat() and compare with
st_rdev) also updated.

	Added old_encode_dev()/old_decode_dev() in dm-ioctl-*.c

	Minor cleanups in md.c + added old_decode_dev() on ioctl argument in
there.

	Minor cleanup of jffs2 fill_super.

	Added missing old_encode_dev() in JFS.

	NFS made dev_t-agnostic.  Aside of minor fixes in debugging printks,
and adding old_encode_dev()/old_decode_dev(), the main part is in handling
of exports with large dev_t.  New fhandle format introduced, fh_verify(),
fh_compose() and exports cache taught to deal with it.  Format is used when
->s_dev of exported fs doesn't fit into 256:256; in that case we put major
and minor in separate words in fhandle; ->fh_fsid_type is set to 2.

	tty->device had been used only in a couple of places and can be
calculated by tty->index and tty->driver.  Field removed, its users switched
to static inline dev_t tty_devnum(tty).

	sanitized dev_t use in debugging printks in XFS, switched their
bhash to use of struct block_device, added old_encode_dev() in XFS statfs()
(it had stored dev_t in ->fs_id[0])

	added old_encode_dev() in drm debugging printks, so they won't change
when we go for bigger dev_t.

	fs/reiserfs/procfs.c made sane.  It used to store dev_t of filesystem
in the proc_dir_entry->data (which is void *) and played very odd games after
that.
	Switched to seq_file, stores pointer to superblock, uses sget() to
validate it, avoids use of dev_t completely.

into kernel.bkbits.net:/home/davem/sparc-2.5

into kernel.bkbits.net:/home/davem/net-2.5